Dataset Viewer
The dataset viewer is not available for this subset.
The chief_10x config contains 31 while it should generally contain 3 splits maximum (train/validation/test). If the splits TCGA_ACC, TCGA_BLCA, TCGA_BRCA, TCGA_CESC, TCGA_CHOL... are not used to differentiate between training and evaluation, please consider defining configs of this dataset instead. You can find how to define configs instead of splits here: https://huggingface.co/docs/hub/datasets-data-files-configuration
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
TCGA Pathology Feature Atlas
This repository stores verified TCGA pathology foundation-model embeddings as sharded Parquet files.
Physical layout:
data/{project_id}/{model_name}/{magnification}/part-*.parquet
Included release scope
- Models: chief, conch_v1, ctranspath, gigapath, gigapath_slide, hibou_l, hoptimus0, titan, uni_v1, uni_v2
- Groups: 764
- Shards: 7,125
- Embedding rows: 5,922,587
- Registry rows: 279,001
- H5 validation rows: 279,001
- H5 validation failures: 0
Each row is a chunk of patch or slide embeddings from one TCGA slide.
Example streaming load:
from datasets import load_dataset
ds = load_dataset("IUCompPath/TRACE", name="conch_v1_20x", split="TCGA_BRCA", streaming=True)
- Downloads last month
- 1,954